On August 15, DeepSeek officially released DeepSeek-V4-Pro in GA (Generally Available), completing the V4 model family’s transition from preview to commercial release. This model has appeared on the app/web via “Expert Mode” and is available via API—with the model name remaining the same and only internal adjustments made.

The first highlight of the release lies in the agent domain. DeepSeek describes V4-Pro as delivering “major agent upgrades” aimed at production efficiency. Accompanying this is a flexible reasoning effort mechanism, applicable to both V4-Pro and V4-Flash, allowing users to select the reasoning level: low for simple tasks, high for daily agent workflows, and max for complex problems requiring deep reasoning.

The second highlight is ecosystem compatibility. V4-Pro supports the native OpenAI Responses API, pre-optimized by DeepSeek for Codex with one-touch setup—a move that helps development teams familiar with OpenAI standards switch over with virtually no code modifications.

Regarding pricing, DeepSeek updated its API pricing and introduced a peak/off-peak mechanism, effective from 16:00 UTC on August 16. Peak hours are 01:00–04:00 and 06:00–10:00 UTC; outside those hours is off-peak, and prices are 50% cheaper. Specifically:

  • DeepSeek-V4-Flash: during peak hours, input (cache miss) is $0.44, cache hit is $0.014, and output is $1.32 per 1 million tokens; during off-peak hours, they are $0.22, $0.007, and $0.66, respectively.
  • DeepSeek-V4-Pro: during peak hours, input is $1.32, cache hit is $0.044, and output is $3.96; during off-peak hours, they are $0.66, $0.022, and $1.98, respectively.

With discounts of up to half, the off-peak mechanism offers the clearest benefit to user groups running high-volume workloads that are not latency-sensitive, typically background agent pipelines, batch data processing, or recurring jobs. Enterprises can shift these types of tasks to off-peak hours to slash their API budget by nearly half without impacting efficiency.

Overall, the V4-Pro release shows DeepSeek simultaneously pursuing two goals: boosting production-grade agent capabilities and making API pricing more flexible for enterprise customers. While the peak/off-peak mechanism is new to DeepSeek, it is increasingly becoming a common trend among many model providers to balance infrastructure loads during peak hours.